Advanced Modeling of Hydrogen Mixture Properties for Production and Transfer Processes Using Machine Learning Techniques

Abstract

This study investigates the use of machine learning (ML) techniques to predict the density of hydrogen-methane binary mixtures over a wide range of temperatures and pressures. Accurate density prediction is critical for optimizing hydrogen production, transportation, and utilization. Traditional methods like experimental measurements and equations of state (EoS) are either costly or computationally intensive. To address these challenges, four ML models—XGBoost, Gradient Boosting Decision Trees (GBDT), Random Forest (RF), and k-Nearest Neighbors (KNN)—were trained on 4,000 density data points generated using the GERG-2008 EoS. The models were evaluated using metrics such as RMSE, MAE, and R2. Among the models, GBDT outperformed others with the lowest RMSE (0.44) and MAE (0.23), achieving an R2 of 1.00. XGBoost followed closely, while RF and KNN showed moderate and lower accuracy, respectively. The results demonstrate that ML-based approaches, particularly GBDT, provide a reliable and efficient alternative to conventional methods for predicting the thermophysical properties of hydrogen mixtures, contributing to the development of a sustainable hydrogen economy.